AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Margin-Adaptive Confidence Ranking for Reliable LLM Judgement

Gaojie Jin, Yong Tao, Lijia Yu, Tianjin Huang

ICML 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add adaptive margin to confidence estimator training (ablation)Learning Confidence (Vanilla), Mistral-7B · AlpacaEvalranking loss RK: 0.3865 -> 0.3393Improved
add adaptive margin to confidence estimator training (ablation)Learning Confidence (Vanilla), Mistral-7B · AlpacaEvalAUROC: 0.6179 -> 0.6672Improved
add adaptive margin (ablation)Learning Confidence (Vanilla), Llama3-70B · AlpacaEvalranking loss RK: 0.3236 -> 0.2776Improved
add adaptive margin (ablation)Learning Confidence (Vanilla), Qwen2.5-72B · AlpacaEvalranking loss RK: 0.3370 -> 0.2707Improved
add adaptive margin (ablation)Learning Confidence (Vanilla), Mistral-7B · HH-RLHFranking loss RK: 0.3841 -> 0.3286Improved
add adaptive margin (ablation)CSE + Vanilla, L->Q->O · AlpacaEvalguarantee success rate (%): 93.4 -> 94.8Improved
add adaptive margin (ablation)CSE + Vanilla, L->Q->O · HH-RLHFguarantee success rate (%): 37.9 -> 51.6Improved
add adaptive margin (ablation)CSE + Vanilla, M->L->O · AlpacaEvalguarantee success rate (%): 75.7 -> 82.1Improved
add adaptive margin (ablation)CSE + Vanilla, L->Q->O · TL;DRguarantee success rate (%): 75.4 -> 79.2Improved
replace Simulated Annotators confidence with learned confidenceCSE, L->Q->O · Chatbot Arenaguarantee success rate (%): 90.2 -> 92.7Improved
replace perfect-calibration GPT-4 heuristic with CSE using learned confidenceHeuristic Selection, L->Q->O · AlpacaEvalguarantee success rate (%): 0.0 -> 94.8Improved
replace Heuristic Selection with CSE + OursHeuristic Selection · Chatbot Arena coverage (L->Q->O)coverage (%): 88.4 -> 56.4Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy